Whisper MCP Server is a Model Context Protocol server that provides local audio transcription using Whisper.
The tool solves the problem of transcribing audio files without sending data to external services. It wraps whisper.cpp to run transcription locally on your machine, supporting multiple model sizes from tiny to large, and handling various audio formats including wav, mp3, and m4a. Transcriptions can be returned as plain text, with timestamps, or as JSON.
This tool suits developers building voice-driven workflows on macOS who want to keep audio processing local and private. It works best for those already using Claude Desktop or Claude Code and who need flexible model selection based on their speed and accuracy requirements. The project explicitly positions itself as a minimal implementation designed to pair with apple-voice-memo-mcp for a complete voice memo workflow, suggesting it fills a specific niche rather than attempting to be a general-purpose transcription solution.
The project shows focused, deliberate development activity. The codebase is written in TypeScript and maintains clear separation of concerns through its MCP tool interface. Documentation is thorough, covering installation, configuration for both Claude Desktop and Claude Code, and practical workflow examples. The tool provides three distinct MCP operations: transcribing audio files with configurable models and output formats, listing available models and their download status, and downloading models on demand. Requirements are straightforward and platform-specific, targeting macOS with Apple Silicon testing and requiring Node.js, whisper.cpp, and ffmpeg as dependencies.
